This subproject is one of many research subprojects utilizing the resources provided by a Center grant funded by NIH/NCRR. The subproject and investigator (PI) may have received primary funding from another NIH source, and thus could be represented in other CRISP entries. The institution listed is for the Center, which is not necessarily the institution for the investigator. A deep scientific understanding of the fundamental biological mechanisms underlying brain function will have a revolutionary impact on national interests in science, medicine, economic growth, security, and human wellbeing. Yet after more than 100 years of modern neuroscience research, we still do not understand many of the fundamental principles by which the brain controls the body, makes plans, and stores episodic memories, much less how the combined activity of millions of neurons leads to consciousness. It is generally accepted that the mammalian brain is the most complicated system in the known universe. It is this complexity that makes understanding the brain difficult. It would be very useful to use detailed large-scale brain models incorporating quantitative descriptions of neuronal physiology, synaptic plasticity, local circuitry, and global anatomy in order to explore, in a bottom-up fashion, the fundamental principles at work in the brain. BACKGROUND Over the past two decades, numerous modeling efforts have been made at The Neurosciences Institute with increasing levels of realism (Edelman, 1987;Reeke and Edelman 1987;Almassy et al. 1998;Seth et al., 2004;Izhikevich, et al., 2004). The most realistic (Izhikevich et al, 2004) was a large-scale model consisting of 100,000 spiking neurons, each of which is capable of reproducing rich firing patterns recorded from real neurons (Izhikevich, 2004). Furthermore, the model included several features critical for any realistic simulation of the brain, including spike-timing dependent synaptic plasticity (STDP), local and global reentrant connections (Edelman, 1987), conduction delays, and short-term synaptic facilitation and depression. CURRENT MODEL Our latest brain simulation is designed to overcome several limitations of our previous models along several important dimensions. While the new model maintains all of the features of our prior work, it also incorporates important new data emerging from the efforts to catalog the detailed microcircuitry of the cerebral cortex (Binzegger et al., 2004), along with more realistic global anatomy between different brain regions, the details of which are available from human brain imaging results. The goal is to have a reasonable representation of the anatomy of thalamo-cortical circuits. The model incorporates a new single-cell model with multiple compartments (Izhikevich, 2007a), along with the effect of the neuromodulator dopamine on synaptic plasticity (Izhikevich, 2007b). Mathematically, the model is a system of delay-differential equations, which we simulate using the forward Euler method with time step equivalent to 0.5 ms of neural activity. For more details of the latest brain model simulation, please see http://vesicle.nsi.edu/users/izhikevich/human_brain_simulation/Blue_Brain.htm. A completely parallel version has been coded in C++ using the standard Message Passing Interface (MPI) library. PROPOSED RESEARCH We plan to explore this new model, in a bottom-up fashion, studying it as a neurophysiologist would do with an animal subject. To be successful, we estimate that we will need to model the activity of at least one million neurons over an hour or more of simulated time. Initially, we will pursue two specific areas: 1. Parameter exploration. Although the neural activity of the model already looks reasonable, we currently have modeled ongoing activity only during sleep. Extensive parameter exploration will be required to find a regime that resembles other well-known activity states of the brain. 2. Measurement of simulated local field potentials, MEG, EEG, and fMRI activity in the model will be conducted for comparison with human data during different behavioral states (e.g., sleep, quiet waking, and epilepsy). Similar work has been carried out by others on simpler models (e.g. Hill and Tononi 2005, Honey et al., 2007). After benchmarking the code on several of our local systems, we have determined that, in order to make progress, we will need to go well beyond our present computing resources;this is not surprising given the ambitious goals of the project. LOCAL COMPUTING ENVIRONMENT The brain simulation software is currently running on a Linux cluster with 30 nodes, each with 2 Intel Pentium Xeon processors running at 3GHz with 2 Gigabytes of shared memory. The cluster is connected with a 32-port Myrinet switch. Not only does it take 10 days to simulate a single hour of brain activity for one million neurons, but also we are constrained by the RAM to the smaller scale models. COMPUTATIONAL NEEDS If we are to make practical progress on this project, we need to have access to much higher capacity computational resources, such as those available through TeraGrid. Thus we are requesting the use of a 1024 node Linux cluster with at least 2 gigabytes of RAM per node, such as the TACC Dell Linux Cluster. The SDSC IA-64 Cluster would be a second possibility. We are requesting the maximum 30,000 service units, which should allow for optimizing and testing the existing code (which we believe will not be a major effort) and for several production runs. If our application to the Development Allocation Committee is approved and the project is successful, we will likely apply for additional time to the Medium Allocation Committee. REFERENCES Almassy, N., Edelman, G.M. and Sporns, O. (1998). Behavioral constraints in the development of neuronal properties: A cortical model embedded in a real-world device, Cereb Cortex 8:346-61. Binzegger T, Douglas RJ, and Martin KAC (2004). A quantitative map of the circuit of cat primary visual cortex. J Neurosci. 24(39):8441-8453. Edelman, GM (1987). Neural Darwinism: The Theory of Neuronal Group Selection. New York: Basic Books, Inc. Hill S, Tononi G. (2005). Modeling sleep and wakefulness in the thalamocortical system. J Neurophysiol. 93(3):1671-98. Honey CJ, et al. (2007). Network structure of cerebral cortex shapes functional connectivity on multiple time scales. Proc. Nat. Acad. Sci. USA 104(24): 10240-10245. Izhikevich EM, Gally JA, and Edelman GM (2004) Spike-Timing Dynamics of Neuronal Groups. Cereb Cortex 14:933-944. Izhikevich EM (2004) Which model to use for cortical spiking neurons? IEEE Transactions on Neural Networks, 15:1063-1070. Izhikevich EM (2007a) Dynamical Systems in Neuroscience: The Geometry of Excitability and Bursting. Cambridge: The MIT Press. Izhikevich EM (2007b). Solving the distal reward problem through linkage of STDP and dopamine signaling. Cereb Cortex 10.1093/cercor/bhl152. Reeke G.N. Jr. and Edelman G.M. (1987) Selective neural networks and their implications for recognition automata, Intl. J. Supercomputer Appl. 1:44-69 (1987). Seth, AK, McKinstry, JL, Edelman, GM and Krichmar, JL (2004) Visual binding through reentrant connectivity and dynamic synchronization in a brain-based device. Cereb Cortex 14:1185-1199.